Intelligent evaluation system for mild cognitive impairment

Abstract

The invention discloses an intelligent evaluation system for mild cognitive impairment, and belongs to the technical field of computer science. The invention discloses an intelligent evaluation system for mild cognitive impairment. The system comprises a user login module, an evaluation module and a historical recording module, wherein the user login module obtains basic information of a target user, wherein the basic information comprises a name, a gender, a birth month, an identity card number, an educated degree and a residential address; the evaluation module provides an electronic Monte Renal cognitive evaluation scale for the target user and performs cognitive evaluation; and the historical recording module provides a test related information display page for the target user, wherein the test related information display page comprises a test name, time, a state and a score. Efficiency and accuracy of cognitive function evaluation are improved through a computer, and a standardized evaluation process is realized through a digital and automatic test process.

Description

Technical Field

[0001] The present invention belongs to the field of computer science and technology, and in particular relates to an intelligent evaluation system for mild cognitive impairment. Background Art

[0002] Mild cognitive impairment (MCI) is a state of mild cognitive decline that lies between normal aging and dementia. People with MCI may experience mild difficulties with memory, attention, language, executive function, or visual-spatial abilities, but these symptoms usually do not significantly affect daily life activities.

[0003] The Montreal Cognitive Assessment (MoCA) is a widely used brief cognitive screening tool for detecting mild cognitive impairment. The scale covers multiple cognitive domains, including attention, executive function, memory, language, visuospatial ability, naming, abstract thinking and orientation, with a total of 11 items; it is designed to identify early Alzheimer's disease and other forms of dementia. Compared with the traditional Mini-Mental State Examination (MMSE), the MoCA scale shows higher sensitivity and specificity in assessing mild cognitive impairment. The MMSE mainly detects a decline in overall cognitive function, but is not sensitive and specific enough for the assessment of mild cognitive impairment. The MoCA scale places more emphasis on the assessment of executive function and attention, which makes it more suitable for screening mild cognitive impairment.

[0004] However, although the paper version of the Montreal Cognitive Assessment is widely used in clinical practice and research, it also has some limitations. For example, manual entry of paper-based scale data is prone to errors, paper records are not easy to store and retrieve, large-scale analysis is difficult, and dynamic monitoring and evaluation of cognitive function changes over time are difficult to achieve. Summary of the invention

[0005] In view of the deficiencies of the prior art, the purpose of the present invention is to provide an intelligent assessment system for mild cognitive impairment, which solves the problems in the prior art.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An intelligent assessment system for mild cognitive impairment, comprising:

[0008] User login module: obtain the target user's basic information, including name, gender, date of birth, education level, ID number and residential address;

[0009] Assessment module: Provide target users with the electronic Montreal Cognitive Assessment Scale and conduct cognitive assessment;

[0010] And, the history module: provides the target users with a test-related information display page, including: test name, time, status and score.

[0011] Furthermore, the evaluation module includes:

[0012] Task units: including alternating line task, cube copying task, clock drawing task, naming task, memory task, attention task, sentence repetition task, word fluency task, abstract task, delayed recall task, and orientation task;

[0013] Human-computer interaction unit: including touch interaction, voice interaction and picture interaction functions, adapted to different tasks of the Montreal Cognitive Assessment Scale;

[0014] Algorithm unit: used to receive and understand the target object's feedback during the cognitive ability assessment process, and score according to the target object's feedback and standards. The algorithm unit includes: speech recognition algorithm, sequential execution algorithm, image evaluation algorithm, keyword extraction algorithm and text matching algorithm;

[0015] And the front-end patient application unit: provides an interactive interface for the target subjects to complete cognitive tests on different devices.

[0016] Furthermore, the touch interaction function is: the target object performs click and slide operations by touching the screen;

[0017] The voice interaction function is used to guide the target subject when taking the test, and to record the target subject's answers when answering voice tasks, so as to test and understand the target subject's understanding and feedback of the questions;

[0018] The picture interaction function is used to enable the target subject to take a picture of the picture drawn on paper and upload it when performing the drawing task, so as to realize the analysis of the picture and the presentation of the picture in the naming task.

[0019] Furthermore, the speech recognition algorithm is used to recognize the target object's speech and transcribe it into text.

[0020] Furthermore, the picture evaluation algorithm comprises the following steps:

[0021] Instruct the target subject to draw pictures taken under various backgrounds, colors, and lighting conditions, and manually score each picture to obtain a picture library;

[0022] Standardize the images, including image size, cropping, and adjustment of contrast, transparency, and brightness to ensure that all images are in the same format;

[0023] For the data in the image library, a multimodal fusion network structure is used; for the extraction of text features, the scoring criteria of the task are regarded as prompt text;

[0024] Select a multimodal pre-trained model based on the transformer architecture for training, and adjust the model parameters by inputting the pre-processed image and text data into the multimodal pre-trained model.

[0025] Furthermore, the keyword extraction algorithm includes a keyword speech recognition model, which recognizes and matches reserved keywords from speech, and the steps include:

[0026] After randomly selecting a ten-second background audio clip, insert 0-4 audio clips of the keyword into this 10-second clip and annotate it;

[0027] Perform sound standardization, silence detection and cropping, windowing and framing, word segmentation, and feature extraction on the speech dataset;

[0028] Select a basic speech recognition model and train it to adjust the model accuracy.

[0029] Furthermore, the text matching algorithm includes:

[0030] Sentence similarity matching algorithm: find the largest overlap between two strings and determine the similarity by calculating the overlap and the total length of the strings;

[0031] Exact matching algorithm; used to compare the target object's answer with the standard answer and give a corresponding score.

[0032] Furthermore, the implementation steps of the sentence similarity matching algorithm include:

[0033] Record the target user's answer and transcribe it into text using speech recognition technology;

[0034] Preprocess the transcribed text to remove irrelevant characters, including spaces and punctuation;

[0035] Find the parts of the transcribed text that are identical to the model answer and determine the largest overlap between the two texts;

[0036] The length of the two texts and the maximum overlap are taken as input and the similarity score is returned. If the similarity is greater than a certain threshold, the answer is considered correct.

[0037] Furthermore, the calculation formula of the similarity score is:

[0038] similarity score=2×LCS length / (s1 length+s2 length)

[0039] Among them, LCS length refers to the maximum overlapping part of two strings, and s1 length and s2 length represent the lengths of the two strings respectively.

[0040] Furthermore, the alternating line-joining task is used to assess executive function and attention, allowing the user to alternately click on numbers and Chinese characters;

[0041] The copy cube task is used to assess visual-spatial ability and executive function, allowing users to imitate and draw according to the given cube picture and click to upload;

[0042] The clock drawing task is used to assess executive function and memory, and allows the user to draw a complete clock of a specified time and click to upload;

[0043] The naming task is used to assess the user's language and vocabulary access ability, allowing the user to name and record three animal pictures;

[0044] The memory task is used to assess immediate memory and allows the user to memorize five cued words;

[0045] The attention tasks described were used to assess attention and concentration;

[0046] The sentence retelling task is used to assess language comprehension and memory, and allows the user to memorize two prompted sentences and record them;

[0047] The word fluency task is used to assess language production and vocabulary access speed, and allows users to say as many words beginning with "yi" as possible within one minute and record them;

[0048] The abstract task is used to assess abstract thinking and concept formation ability, and allows users to identify the relationship between two words and record the audio;

[0049] The delayed recall task is used to assess short-term memory retention and allows users to recall five previously mentioned words without prompts and record them;

[0050] The orientation task is used to assess orientation and allows the user to provide information about time and place and record an audio recording.

[0051] A device for detecting mild cognitive impairment comprises the above-mentioned intelligent assessment system for mild cognitive impairment.

[0052] Beneficial effects of the present invention:

[0053] 1. The intelligent assessment system developed by the present invention based on the "Twelfth Five-Year Plan" version of the Montreal Cognitive Assessment Scale aims to improve the efficiency and accuracy of cognitive function assessment through technical means. This system realizes a standardized assessment process through a digital and automated testing process.

[0054] 2. The present invention can guide users through tests in various cognitive areas, including memory, attention, executive function, language, abstract thinking, orientation and visual-spatial ability; the system automatically records the user's response and completion time, reducing the need for manual intervention and ensuring the consistency and repeatability of the evaluation process.

[0055] 3. The evaluation system of the present invention also has the ability to process and analyze data, and can quickly generate scores. This not only speeds up the evaluation, but also improves the objectivity of the results, because the system reduces the possibility of human error and bias. Through precise algorithms, the system can conduct a detailed analysis of the user's cognitive function, thereby providing more accurate evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 It is a schematic diagram of the evaluation process of the intelligent evaluation system for mild cognitive impairment of the present invention;

[0058] Figure 2 This is a diagram of the information entry interface of the user login module of the present invention;

[0059] Figure 3 This is the interface diagram of the evaluation module of the present invention;

[0060] Figure 4 This is a diagram of the final score feedback interface in the front-end patient application unit of the present invention;

[0061] Figure 5 It is a diagram of the interface of the history record module of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] An intelligent assessment system for mild cognitive impairment includes: a user login module, an assessment module, and a history module. The assessment process is as follows: Figure 1 As shown; the user first authorizes login through the user login module and fills in personal information such as name, gender, age, etc., and then can complete the test through the assessment module and obtain a cognitive ability score. After completing the assessment, the historical assessment results can be viewed through the history module.

[0064] User login module

[0065] like Figure 2 As shown, the user login module is used to obtain the basic information of the target user, including name, gender, date of birth, education level, ID number, residential address and other basic information, which can be used for subsequent research.

[0066] When in use, the target user's family members are required to click on the user login module and fill in the target user's basic information one by one to ensure the accuracy and completeness of the collected information.

[0067] Evaluation Module

[0068] like Figure 3 As shown, the evaluation module includes: human-computer interaction module, algorithm unit and front-end patient application unit;

[0069] 1) Task Unit

[0070] The task unit refers to the "Montreal Cognitive Assessment Test Scale", and its tasks include: alternating line task, cube copying task, clock drawing task, naming task, memory task, attention task, sentence repetition task, word fluency task, abstract task, delayed recall task, and orientation task.

[0071] The alternating connection task is used to assess executive function and attention, and allows users to alternately click to connect numbers and Chinese characters;

[0072] The copy cube task is used to assess visual-spatial ability and executive function, allowing users to imitate and draw according to the given cube picture and click to upload;

[0073] The clock drawing task is used to assess executive function and memory, and allows the user to draw a clock with a specified time and click to upload;

[0074] The naming task is used to assess the user's language and vocabulary access ability, allowing the user to name and record three animal pictures;

[0075] The memory task is used to assess immediate memory and allows the user to memorize five cued words;

[0076] The attention task is used to assess attention and concentration, and allows the user to complete three small tests in sequence: repeating numbers, clicking to "1", and subtracting 7 from 100 continuously;

[0077] The sentence retelling task is used to assess language comprehension and memory, and allows the user to memorize two prompted sentences and record them;

[0078] The word fluency task is used to assess language production and vocabulary access speed, and allows users to say as many words beginning with "yi" as possible within one minute and record them;

[0079] The abstract task is used to assess abstract thinking and concept formation ability, and allows users to identify the relationship between two words and record the audio;

[0080] The delayed recall task is used to assess short-term memory retention and allows users to recall five previously mentioned words without prompts and record them;

[0081] The orientation task is used to assess orientation and allows the user to provide information about time and place and record an audio recording.

[0082] 2) Human-computer interaction unit

[0083] The human-computer interaction module is used to receive the user's operating instructions, thereby realizing the automated management of the cognitive assessment process. This unit can be adapted and optimized according to the specific needs of different question types to ensure the smoothness and accuracy of the assessment process;

[0084] The human-computer interaction unit has the following functions: touch interaction, voice interaction and picture interaction. These functions are adapted to the different tasks of the Montreal Cognitive Assessment Scale to maximize the automation of the scale.

[0085] 2.1) The touch interaction function is an operation of clicking, sliding, etc. of the target object by touching the screen; for the touch interaction operation, the specific implementation process of the present invention is as follows:

[0086] (1) After completing each question, the target subject uploads the answer and jumps to the next task by clicking the “Next” button on the screen;

[0087] (2) After completing the drawing task, the target object uploads the picture to the backend by clicking “Upload Picture”;

[0088] (3) The alertness test requires the target subject to respond when hearing the system read out the number 1. When conducting this test, the present invention requires the target subject to click the "click here" button on the screen when hearing the number 1 as the target subject's alertness response.

[0089] 2.2) The voice interaction function is used to guide the target subject when taking the test and record the target subject's answers when answering voice tasks, so as to test and understand the target subject's understanding and feedback on the questions;

[0090] This function takes into account that most of the target objects are elderly people over 65 years old. For the elderly, children's voices are obviously more friendly. Therefore, the present invention integrates the voice interaction of virtual human technology and uses children's voices to interact with the target objects, so that the target objects can be more immersed and relaxed during the cognitive function assessment process.

[0091] 2.3) The picture interaction function is used for the target subject to take a picture of the picture drawn on paper and upload it when doing the drawing task, so as to realize the picture analysis and the picture presentation in the naming task;

[0092] The "Montreal Cognitive Assessment Test Scale" requires the target subject to draw a cube and a clock, but the elderly often have less exposure to electronic devices and have poor hand stability when drawing, which makes accurate drawing of pictures on the screen a major difficulty in realizing the electronicization of the scale. For this reason, the present invention adopts the traditional drawing form and requires the target subject to take a photo and upload it after completing the drawing, which can well avoid the inconvenience brought by the electronic brush.

[0093] The "Montreal Cognitive Assessment Test Scale" includes a picture naming task. For this purpose, the present invention presents common and vivid animal pictures to the target object to increase the interest of the target object test and facilitate a more in-depth assessment of the user's cognitive level.

[0094] 3) The algorithm unit is used to receive and understand the target object's feedback during the cognitive ability assessment process, and to score according to the target object's feedback combined with professional standards;

[0095] The algorithm units include: speech recognition algorithm, sequential execution algorithm, image evaluation algorithm, keyword extraction algorithm and text matching algorithm;

[0096] 3.1) The speech recognition algorithm is used to recognize the speech of the target object and transcribe it into text for subsequent analysis. Specifically, the present invention adopts a general speech recognition model (including but not limited to whisper), which is trained based on a large amount of labeled audio, so that it can have higher accuracy and robustness in Chinese speech recognition compared to other speech recognition methods, and has higher recognition ability for accents, background noise and technical terms, and is particularly suitable for transcribing long texts.

[0097] For example, for the fluency test task in the Montreal Cognitive Assessment Test Scale, the target subject is required to say as many words as possible that begin with the homophone "yi" within 1 minute. Therefore, after receiving the target subject's answer, the speech recognition algorithm converts the word segmentation into pinyin for subsequent analysis.

[0098] 3.2) Sequential execution algorithm is a basic algorithm design pattern, which executes a series of operations or steps in a certain order. This algorithm is simple and intuitive, easy to understand, each step is a direct successor to the previous step, there is no jump or concurrent execution, and it is particularly suitable for situations where tasks need to be completed strictly in a specific order;

[0099] Taking the cross-connection task in the Montreal Cognitive Assessment Test Scale as an example, when performing the cross-connection task, the target object is required to connect in the order of "1-A-2-B-3-C-4-D-5-E". The present invention uses a sequential execution algorithm to force the target object to click the corresponding buttons on the screen in order. If the button is clicked incorrectly, no points will be given.

[0100] 3.3) Image evaluation algorithms include training multimodal models and prompt (In-context-learning) technology;

[0101] The image evaluation algorithm aims to accurately quantify the performance of the "copy cube" and "draw a clock" tasks in the scale. Specifically, considering that the target object may face different painting backgrounds, different brush types, and different lighting conditions when performing painting tasks, the present invention trains a multimodal evaluation model based on prompt technology, which has strong feature extraction and representation capabilities, can adapt to diverse painting environments and make more accurate scores. The image evaluation algorithm includes the following steps:

[0102] Step 1: Dataset construction: Invite a group with a balanced gender and age distribution, and instruct them to draw pictures taken under various backgrounds, colors, and different lighting conditions, including but not limited to standard cubes or standard clocks, and manually score each picture to obtain a picture library;

[0103] Step 2: Image preprocessing: Use image processing libraries (including but not limited to PIL) to standardize images, including image resizing, cropping, and adjustment of contrast, transparency, and brightness to ensure that all images are in the same format for easy model processing and analysis.

[0104] Step 3, model architecture: The present invention adopts a multimodal fusion network structure for the data in the image library. In particular, for the extraction of text features, the scoring criteria of the task are regarded as a specific prompt text, so as to train a multimodal model that integrates images and text. The following is an example of the "drawing a clock" task:

[0105]

[0106] Step 4, model training: The present invention selects a multimodal pre-trained model based on the transformer architecture for fine-tuning training. The pre-processed image and text data are input into the pre-trained model to adjust the model parameters and optimize the model's adaptability to the scale scoring criteria.

[0107] 3.4) Keyword extraction algorithms are mainly used in short text tasks such as "word similarity", "word delayed recall", "picture naming", "number forward and backward recitation" and "orientation test". Since speech recognition technology has poor accuracy in transcribing short texts, the present invention considers using a multimodal speech recognition model to score the target object's answer. The multimodal speech recognition model considers both audio and text features and can better combine context to make judgments. The keyword extraction algorithm of the present invention includes: a keyword speech recognition model and prompt technology.

[0108] The keyword speech recognition model is to recognize and match reserved keywords from speech, and the specific steps are as follows:

[0109] The first step is to construct a speech data set: randomly select a 10-second background audio clip, insert 0-4 audio clips of the keyword into this 10-second clip, and mark it to make the start and end time of the keyword clear; in addition, it is necessary to record some words unrelated to the keyword and perform the same operation;

[0110] The second step is to preprocess the speech data: perform sound standardization, silence detection and cropping, windowing and framing, word segmentation, and feature extraction on the speech data set;

[0111] The third step is to train the keyword recognition model: select a basic speech recognition model and train it to adjust the model accuracy and optimize the model's accuracy in keyword recognition.

[0112] The keyword speech recognition model is used to extract speech segments related to cognitive assessment tests from the target object's complex speech for subsequent scoring. It is particularly suitable for tasks such as "picture naming" and "delayed memory of words" in the scale.

[0113] The extracted relevant speech segments are input into the multimodal model, and the prompt technology is used to set specific prompts or guiding sentences to enable the model to give a more accurate score. When completing the scale task, this embodiment designs different prompts for different question types. The following is an example of the naming of "lion" in the "picture naming" task:

[0114]

[0115]

[0116] Using a multimodal model based on text and audio features to evaluate user responses and comprehensively considering audio and text features can effectively avoid the problem of large errors in speech recognition models when transcribing shorter speech due to issues such as accents, thereby improving the credibility of the results to a certain extent.

[0117] 3.5) Text matching algorithms include: sentence similarity matching algorithm and complete matching algorithm;

[0118] The similarity matching of sentences uses the Ratcliff / Obershelp algorithm. The core idea of ​​the algorithm is to find the largest overlapping part between two strings, and then determine the similarity by calculating the overlapping part and the total length of the string. It is especially suitable for the "sentence retelling" task to improve the accuracy of the task. The implementation steps are as follows:

[0119] First, the target user’s answering voice is recorded and transcribed into text using speech recognition technology;

[0120] The transcribed text is then preprocessed to remove irrelevant characters such as spaces and punctuation marks;

[0121] Then the algorithm is used to find the parts of the transcribed text that are identical to the standard answer, and by comparing these parts, the maximum overlap between the two texts is determined;

[0122] Finally, the length of the two texts and the maximum overlap are used as input, and the similarity score is returned. If the similarity is greater than a certain threshold, the answer is considered correct; the calculation formula for the similarity score is:

[0123] similarity score=2×LCS length / (s1 length+s2 length)

[0124] Among them, LCS length refers to the maximum overlapping part of two strings, and s1 length and s2 length represent the lengths of the two strings respectively.

[0125] The complete matching algorithm is used to compare the target object's answer with the standard answer and give a corresponding score; the complete matching algorithm is particularly suitable for questions with standard answers, such as the "calculation combination" task in the scale.

[0126] In addition, the present invention takes into account the time factor and completes the "alertness test" task by recording the time difference. Specifically, the "alertness test" requires the evaluator to read out a string of numbers and requires the target object to respond when hearing the number 1. Based on this, the present invention performs this evaluation by recording the time difference and number of times the target object responds to the number 1 in the prepared audio.

[0127] Finally, the Montreal Cognitive Assessment has high requirements for educational level. Therefore, in addition to the above tasks, the present invention considers the educational level factor based on clinical verification, with 12 years as the boundary. If the education level is less than 12 years, one point is added to the total score to improve the accuracy of the assessment.

[0128] 4) The front-end patient application unit is designed to provide an intuitive and easy-to-use interactive interface for the target subjects so that they can complete the Montreal Cognitive Test on different devices. The interface contains clear prompts to help the target user complete basic information entry, complete the assessment, and add up the scores after the target user completes all the tests, and feedback to the user's operation, such as feedback results Figure 4 In addition, after the target object completes the evaluation task, the present invention can automatically collect relevant data, including paintings, voice recordings, and text inputs, and encrypt and transmit these data to the back-end server for further analysis.

[0129] History Module

[0130] like Figure 5 As shown in , the history module is used to store the target object operation information. Figure 5 As shown in the figure, this module provides users with a comprehensive information display page, which contains a number of key test-related information, such as test name, time, status and score, which is convenient for users to track and review; the specific timestamp of each operation is recorded in detail, accurate to microseconds, so as to accurately track the time when the operation occurred. In addition, this module provides a timeline view, and the target object can view the operation history in chronological order, which is convenient for analyzing the time pattern and trend of the operation.

[0131] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0132] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An intelligent assessment system for mild cognitive impairment, characterized in that: include: User login module: obtain the target user's basic information, including name, gender, date of birth, education level, ID number and residential address; Assessment module: Provide target users with the electronic Montreal Cognitive Assessment Scale and conduct cognitive assessment; And, the history module: provides the target users with a test-related information display page, including: test name, time, status and score.

2. The intelligent assessment system for mild cognitive impairment according to claim 1, characterized in that: The evaluation module includes: Task units: including alternating line task, cube copying task, clock drawing task, naming task, memory task, attention task, sentence repetition task, word fluency task, abstract task, delayed recall task, and orientation task; Human-computer interaction unit: including touch interaction, voice interaction and picture interaction functions, adapted to different tasks of the Montreal Cognitive Assessment Scale; Algorithm unit: used to receive and understand the target object's feedback during the cognitive ability assessment process, and score according to the target object's feedback and standards. The algorithm unit includes: speech recognition model, sequential execution algorithm, image evaluation algorithm, keyword extraction algorithm and text matching algorithm; And the front-end patient application unit: provides an interactive interface for the target subjects to complete cognitive tests on different devices.

3. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The touch interaction function is: the target object clicks and slides by touching the screen; The voice interaction function is used to guide the target subject when taking the test, and to record the target subject's answers when answering voice tasks, so as to test and understand the target subject's understanding and feedback of the questions; The picture interaction function is used to enable the target subject to take a picture of the picture drawn on paper and upload it when performing the drawing task, so as to realize the analysis of the picture and the presentation of the picture in the naming task.

4. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The speech recognition model is used to recognize the speech of the target object and transcribe it into text.

5. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The picture evaluation algorithm comprises the following steps: Instruct the target subject to draw pictures taken under various backgrounds, colors, and lighting conditions, and manually score each picture to obtain a picture library; Standardize the images, including image size, cropping, and adjustment of contrast, transparency, and brightness to ensure that all images are in the same format; For the data in the image library, a multimodal fusion network structure is used; for the extraction of text features, the scoring criteria of the task are regarded as prompt text; Select a multimodal pre-trained model based on the transformer architecture for training, and adjust the model parameters by inputting the pre-processed image and text data into the multimodal pre-trained model.

6. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The keyword extraction algorithm includes a keyword speech recognition model, which recognizes and matches reserved keywords from speech, and the steps include: After randomly selecting a ten-second background audio clip, insert 0-4 audio clips of the keyword into this 10-second clip and annotate it; Perform sound standardization, silence detection and cropping, windowing and framing, word segmentation, and feature extraction on the speech dataset; Select a basic speech recognition model and train it to adjust the model accuracy.

7. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The text matching algorithm includes: Sentence similarity matching algorithm: find the largest overlap between two strings and determine the similarity by calculating the overlap and the total length of the strings; Exact matching algorithm; used to compare the target object's answer with the standard answer and give a corresponding score.

8. The intelligent assessment system for mild cognitive impairment according to claim 7, characterized in that: The implementation steps of the sentence similarity matching algorithm include: Record the target user's answer and transcribe it into text using speech recognition technology; Preprocess the transcribed text to remove irrelevant characters, including spaces and punctuation; Find the parts of the transcribed text that are identical to the model answer and determine the largest overlap between the two texts; The length of the two texts and the maximum overlap are taken as input and the similarity score is returned. If the similarity is greater than a certain threshold, the answer is considered correct.

9. The intelligent assessment system for mild cognitive impairment according to claim 8, characterized in that: The calculation formula of the similarity score is: similarity score=2×LCS length / (s1 length+s2 length) Among them, LCS length refers to the maximum overlapping part of two strings, and s1 length and s2 length represent the lengths of the two strings respectively.

10. The intelligent assessment system for mild cognitive impairment according to claim 2, characterized in that: The alternating line-joining task is used to assess executive function and attention, and allows the user to alternately click on numbers and Chinese characters; The copy cube task is used to assess visual-spatial ability and executive function, allowing users to imitate and draw according to the given cube picture and click to upload; The clock drawing task is used to assess executive function and memory, and allows the user to draw a complete clock of a specified time and click to upload; The naming task is used to assess the user's language and vocabulary access ability, allowing the user to name and record three animal pictures; The memory task is used to assess immediate memory and allows the user to memorize five cued words; The attention tasks described were used to assess attention and concentration; The sentence retelling task is used to assess language comprehension and memory, and allows the user to memorize two prompted sentences and record them; The word fluency task is used to assess language production and vocabulary access speed, allowing users to say as many words beginning with "yi" as possible within one minute and record them; The abstract task is used to assess abstract thinking and concept formation ability, allowing users to identify the relationship between two words and record the audio; The delayed recall task is used to assess short-term memory retention and allows users to recall five previously mentioned words without prompts and record them; The orientation task is used to assess orientation and allows the user to provide information about time and place and record an audio recording.

11. A device for detecting mild cognitive impairment, characterized in that: An intelligent assessment system for mild cognitive impairment comprising the method described in any one of claims 1 to 10.

Images